Automatic detection and characterization of cancer are crucial for initiating early treatment. The majority of artificial intelligence (AI) models designed to detect cancer rely on datasets that are either small or moderate in size and typically focus on a single type of cancer and/or radiotracer. This limitation is a significant bottleneck in the existing training and evaluation methods used for AI in medical imaging and radiology. Now, a novel AI method has been shown to accurately identify six different types of cancer in whole-body PET/CT scans. This tool also automatically quantifies tumor burden, which can help in assessing patient risk, predicting responses to treatment, and estimating survival probabilities.

At the Johns Hopkins University School of Medicine (Baltimore, MD, USA), researchers have developed a deep transfer learning technique (a form of AI) for the fully automated segmentation of tumors and prognosis using whole-body PET/CT scans. The study analyzed data from 611 FDG PET/CT scans of patients with lung cancer, melanoma, lymphoma, head and neck cancer, and breast cancer, in addition to 408 PSMA PET/CT scans from prostate cancer patients. This AI method automatically extracted radiomic features and whole-body imaging metrics from the predicted tumor segmentations to quantify molecular tumor burden and uptake across all studied cancer types.

These quantitative features and imaging metrics were then utilized to construct predictive models that proved to be useful for risk stratification, estimating survival, and predicting treatment response in cancer patients. The researchers expect that in the near future, generalizable and fully automated AI tools will significantly contribute in imaging centers by supporting physicians in the interpretation of PET/CT scans for cancer patients. Furthermore, this deep learning approach could unveil significant molecular insights into the biological processes that are currently under-researched in large patient cohorts.

“In addition to performing cancer prognosis, the approach provides a framework that will help improve patient outcomes and survival by identifying robust predictive biomarkers, characterizing tumor subtypes, and enabling the early detection and treatment of cancer,” said Kevin H. Leung, PhD, research associate at Johns Hopkins University School of Medicine. “The approach may also assist in the early management of patients with advanced, end-stage disease by identifying appropriate treatment regimens and predicting response to therapies, such as radiopharmaceutical therapy.” The study's findings were presented at the 2024 Annual Meeting of the Society of Nuclear Medicine and Molecular Imaging (SNMI). 

http://www.gzjiayumed.com/en/index.asp .